A5043458817- Parallel Computing and Optimization Techniques
- Distributed and Parallel Computing Systems
- Advanced Data Storage Technologies
- Cloud Computing and Resource Management
- Advanced Memory and Neural Computing
- Neural Networks and Reservoir Computing
- Graph Theory and Algorithms
- Semiconductor Quantum Structures and Devices
- Ferroelectric and Negative Capacitance Devices
- Quantum and electron transport phenomena
- Physics of Superconductivity and Magnetism
- Semiconductor materials and devices
- Distributed systems and fault tolerance
- Underwater Acoustics Research
- Advanced Graph Neural Networks
- Embedded Systems Design Techniques
- Software-Defined Networks and 5G
- Photonic and Optical Devices
- Neural dynamics and brain function
- Underwater Vehicles and Communication Systems
- Spectroscopy and Laser Applications
- Software System Performance and Reliability
- Quantum Computing Algorithms and Architecture
- Indoor and Outdoor Localization Technologies
- Simulation Techniques and Applications
Oak Ridge National Laboratory
2011-2022
Quantum Science Center
2017
Advanced Engineering (Czechia)
2010
Georgia Institute of Technology
2000-2003
Case Western Reserve University
1996
Advances in commercial cloud computing necessitate continual evaluation of the cloud's performance on a variety applications. This work looks at compute oriented instances from Amazon Web Services and Microsoft Azure platforms evaluates them with several high-performance benchmarks, including HPCC HPCG. These benchmarks illustrate that most cost competitive solution depends application to be run.
GPUs have become an essential component for building compute clusters with high density and performance per watt. As such scale to 1000s of GPUs, efficiently moving data between the becomes imperative get maximum performance. NVSHMEM is implementation OpenSHMEM standard NVIDIA GPU which allows communication be issued from inside kernels. In earlier work, we shown how can used achieve better application on connected through PCIe or NVLink. part this effort, implement IB verbs Mellanox...
File transfers over dedicated connections, supported by large parallel file systems, have become increasingly important in high-performance computing and big data workflows. It remains a challenge to achieve peak rates for such due the complexities of I/O, host, network transport subsystems, equally importantly, their interactions. We present extensive measurements disk-to-disk using Lustre XFS systems mounted on multi-core servers suite 10 Gbps emulated connections with 0-366 ms round trip...
Novel uses of graphical processing units for accelerated computation revolutionized the field high-performance scientific computing by providing specialized workflows tailored to algorithmic requirements. As era Moore’s law draws a close, many new non–von Neumann processors are emerging as potential computational accelerators, including those based on principles neuromorphic computing, tensor algebra, and quantum information. While development these is continuing mature, impact anticipated...
We study memory states of a circuit consisting small inductively coupled Josephson junction array and introduce basic (write, read, reset) operations logics the circuit. The presented operation paradigm is fundamentally different from conventional single quantum flux logics. calculate stability diagrams zero-voltage outline also access times energies for operations.
We present results related to the performance of an algorithm for community detection which incorporates event-driven computation. define a mapping takes graph 𝒢 system symmetrically connected, spiking neurons and use spike train similarities identify vertex communities. On random with 128 vertices known structure we show how our approach can be used individual communities from neuron responses.
The use of graphical processors for distributed computation revolutionized the field high performance scientific computing. As Moore's Law era computing draws to a close, development non-Von Neumann systems: neuromorphic processing units, and quantum annealers; again are redefining new territory computational methods. While these technologies still in their nascent stages, we discuss potential advance two domains: machine learning, solving constraint satisfaction problems. Each utilize...
In this paper, experimental verification for a cryogenic memory cell consisting of three inductively coupled Josephson junctions is presented. Design and operational logic type was recently introduced. The basic operations (read, write, reset) can be implemented on the same simple circuit fundamentally different from existing single flux quantum-based cells. Here, we present design principles validation write readout operation four versions design, each with parameters. Our results show...
We develop a machine learning (ML) based design space exploration (DSE) method that builds predictive models for various responses of hybrid main-memory system. To overcome the challenges associated with latency, capacity, and power memory systems in future extreme-scale machines, architectures are being considered which novel non-volatile (NVM) augment traditional DRAM. However, way before their actual implementation, these emerging need to be simulated analyzed fully understand...
Titan was the flagship supercomputer at Oak Ridge Leadership Computing Facility (OLCF). It deployed in late 2012, became fastest world and retired on August 2, 2019. With Titan's mission complete, this paper provides a first-order examination of usage its critical resources (CPU, Memory, GPU, I/O) over five-year production period (2015-2019). In particular, we show quantitatively that majority CPU time spent large-scale jobs, which is consistent with policy driving ground-breaking science...
In this paper we analyze a cryogenic memory cell circuit based on small coupled array of Josephson junctions. All the basic operations (e.g., write, read, and reset) are implemented same different junctions in can principle be utilized for these operations. The presented operation paradigm is fundamentally from conventional single quantum flux logics (SFQ). As an example, demonstrate driven by SFQ pulse employing inductively three We have chosen realistic junction parameters state-of-the-art...
Importance of achieving high performance Fourier transforms for Cognitive Radio applications can not be over-emphasized. This includes signal detection in the presence noise power uncertainty, multi-resolution spectrum sensing, minimization subcarriers' side lobes OFDM modulators, multi-str
High Performance Computing has been a driving force behind important tasks such as scientific discovery and deep learning. It tends to achieve performance through greater concurrency heterogeneity, where the underlying complexity of richer topologies is managed software abstraction.In this paper, we present our initial assessment NVSHMEM, an experimental programming library that supports Partitioned Global Address Space model for NVIDIA GPU clusters. NVSHMEM offers several concrete...
Recently, there has been a growing interest in enabling fast data analytics by leveraging system capabilities from large-scale high-performance computing (HPC) systems. OpenSHMEM is popular run-time on HPC systems that used for compute-intensive scientific applications. In this paper, we propose to leverage design distributed in-memory key-value store analytics. Accordingly, have developed SHMEMCache top of its symmetric global memory, efficient one-sided communication operations and general...
We consider a System of Systems (SoS) wherein each system S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</sub> , i = 1, 2, . N, is composed discrete cyber and physical components which can be attacked reinforced. characterize the disruptions using aggregate failure correlation functions given by conditional probability SoS an individual system. formulate problem ensuring survival as game between attacker provider, with utility function term...
Accurate component-level power measurements are nowadays essential for the design and optimization of high-performance computing (HPC) systems applications. Particularly, as more heterogeneous HPC developed, characterizations GPU profiles have become extremely crucial because, although GPUs provide exceptional performance, they do consume substantial amounts power. Currently, there various profiling mechanisms available; however, is no standard way to assess quality such schemes. To address...
The worth of completing parallel tasks is modeled using utility functions, which monotonically-decrease with time and represent the importance urgency a task. These functions define earned by task at its completion. performance such system measured as total all completed over some interval (e.g., 24 hours). To maximize when scheduling dynamically arriving onto high computing (HPC) that oversubscribed energy-constrained, we have designed, analyzed, compared different heuristic techniques....
Scientific and big data computations are increasingly being distributed across wide-area networks, they often require access to remote files. The file systems that directly mounted over networks transparently support such computations, also obviate the need for special purpose transfer tools. In typical systems, is limited local sites, in particular, reach of Lustre system implemented InfiniBand (IB) at most tens miles due 2.5ms latency bound. We describe LNet router methods connect IB...